@@ -1,3 +1,12 @@
2+{-# LANGUAGE DeriveDataTypeable #-}
3+{-# LANGUAGE ScopedTypeVariables #-}
4+{-# LANGUAGE RankNTypes #-}
5+{-# LANGUAGE BangPatterns #-}
6+{-# LANGUAGE GeneralizedNewtypeDeriving #-}
7+{-# LANGUAGE RecordWildCards #-}
8+{-# LANGUAGE MagicHash #-}
1-- | Local nodes
10-- | Local nodes
3module Control.Distributed.Process.Node
12module Control.Distributed.Process.Node
@@ -12,10 +21,6 @@ module Control.Distributed.Process.Node
13-- TODO: Calls to 'sendBinary' and co (by the NC) may stall the node controller.
22-- TODO: Calls to 'sendBinary' and co (by the NC) may stall the node controller.
15−#if ! MIN_VERSION_base(4,6,0)
16−import Prelude hiding (catch)
19import System.IO (fixIO, hPutStrLn, stderr)
24import System.IO (fixIO, hPutStrLn, stderr)
20import System.Mem.Weak (Weak, deRefWeak)
25import System.Mem.Weak (Weak, deRefWeak)
21import qualified Data.ByteString.Lazy as BSL (fromChunks)
26import qualified Data.ByteString.Lazy as BSL (fromChunks)
@@ -23,46 +28,67 @@ import Data.Binary (decode)
23import Data.Map (Map)
28import Data.Map (Map)
24import qualified Data.Map as Map
29import qualified Data.Map as Map
27 , fromList
33 , fromList
29 , partitionWithKey
37 , partitionWithKey
31 , filterWithKey
40 , filterWithKey
32 , foldlWithKey
41 , foldlWithKey
43+import Data.Time.Format (formatTime)
44+#if MIN_VERSION_time(1,5,0)
45+import Data.Time.Format (defaultTimeLocale)
47+import System.Locale (defaultTimeLocale)
34import Data.Set (Set)
49import Data.Set (Set)
35import qualified Data.Set as Set
50import qualified Data.Set as Set
42import Data.Foldable (forM_)
59import Data.Foldable (forM_)
43import Data.Maybe (isJust, fromJust, isNothing, catMaybes)
60import Data.Maybe (isJust, fromJust, isNothing, catMaybes)
44import Data.Typeable (Typeable)
61import Data.Typeable (Typeable)
45import Control.Category ((>>>))
62import Control.Category ((>>>))
46−import Control.Applicative ((<$>))
63+import Control.Applicative
47−import Control.Monad (void, when)
64+import Control.Monad (void, when, join)
48import Control.Monad.IO.Class (MonadIO, liftIO)
65import Control.Monad.IO.Class (MonadIO, liftIO)
49import Control.Monad.State.Strict (MonadState, StateT, evalStateT, gets)
66import Control.Monad.State.Strict (MonadState, StateT, evalStateT, gets)
50import qualified Control.Monad.State.Strict as StateT (get, put)
67import qualified Control.Monad.State.Strict as StateT (get, put)
51import Control.Monad.Reader (MonadReader, ReaderT, runReaderT, ask)
68import Control.Monad.Reader (MonadReader, ReaderT, runReaderT, ask)
52import Control.Exception
69import Control.Exception
54− , AsyncException(ThreadKilled)
55 , SomeException
71 , SomeException
56 , Exception
72 , Exception
74+ , uninterruptibleMask_
59−import qualified Control.Exception as Exception (catch, finally)
78+import qualified Control.Exception as Exception
60−import Control.Concurrent (forkIO, myThreadId)
84+import Control.Concurrent (forkIO, killThread)
85+import Control.Distributed.Process.Internal.BiMultiMap (BiMultiMap)
86+import qualified Control.Distributed.Process.Internal.BiMultiMap as BiMultiMap
61import Control.Distributed.Process.Internal.StrictMVar
87import Control.Distributed.Process.Internal.StrictMVar
63 , withMVar
89 , withMVar
64 , modifyMVar
91 , modifyMVar
66 , newEmptyMVar
92 , newEmptyMVar
68 , takeMVar
94 , takeMVar
@@ -77,6 +103,7 @@ import Control.Distributed.Process.Internal.CQueue
77 , enqueue
103 , enqueue
78 , newCQueue
104 , newCQueue
79 , mkWeakCQueue
105 , mkWeakCQueue
81import qualified Network.Transport as NT
108import qualified Network.Transport as NT
82 ( Transport
109 ( Transport
@@ -88,8 +115,8 @@ import qualified Network.Transport as NT
88 , TransportError(..)
115 , TransportError(..)
89 , address
116 , address
90 , closeEndPoint
117 , closeEndPoint
92 , Connection
118 , Connection
94 , EndPointAddress
121 , EndPointAddress
95 , Reliability(ReliableOrdered)
122 , Reliability(ReliableOrdered)
@@ -106,8 +133,11 @@ import Control.Distributed.Process.Internal.Types
106 , LocalProcessId(..)
133 , LocalProcessId(..)
107 , ProcessId(..)
134 , ProcessId(..)
108 , LocalNode(..)
135 , LocalNode(..)
109− , Tracer(..)
136+ , MxEventBus(..)
110 , LocalNodeState(..)
137 , LocalNodeState(..)
138+ , ValidLocalNodeState(..)
139+ , withValidLocalState
140+ , modifyValidLocalState
111 , LocalProcess(..)
141 , LocalProcess(..)
112 , LocalProcessState(..)
142 , LocalProcessState(..)
113 , Process(..)
143 , Process(..)
@@ -117,9 +147,11 @@ import Control.Distributed.Process.Internal.Types
117 , localPidCounter
147 , localPidCounter
118 , localPidUnique
148 , localPidUnique
119 , localProcessWithId
149 , localProcessWithId
120 , localConnections
151 , localConnections
121 , forever'
152 , forever'
122 , MonitorRef(..)
153 , MonitorRef(..)
154+ , NodeClosedException(..)
123 , ProcessMonitorNotification(..)
155 , ProcessMonitorNotification(..)
124 , NodeMonitorNotification(..)
156 , NodeMonitorNotification(..)
125 , PortMonitorNotification(..)
157 , PortMonitorNotification(..)
@@ -139,54 +171,74 @@ import Control.Distributed.Process.Internal.Types
140 , ProcessInfo(..)
172 , ProcessInfo(..)
141 , ProcessInfoNone(..)
173 , ProcessInfoNone(..)
142 , SendPortId(..)
175 , SendPortId(..)
143 , typedChannelWithId
176 , typedChannelWithId
144 , RegisterReply(..)
177 , RegisterReply(..)
145 , WhereIsReply(..)
178 , WhereIsReply(..)
146 , payloadToMessage
179 , payloadToMessage
147 , createUnencodedMessage
180 , createUnencodedMessage
181+ , unsafeCreateUnencodedMessage
148 , runLocalProcess
182 , runLocalProcess
149 , firstNonReservedProcessId
183 , firstNonReservedProcessId
150 , ImplicitReconnect(WithImplicitReconnect)
184 , ImplicitReconnect(WithImplicitReconnect)
186+import Control.Distributed.Process.Management.Internal.Agent
189+import Control.Distributed.Process.Management.Internal.Types
152−import qualified Control.Distributed.Process.Internal.Trace.Remote as Trace
192+import qualified Control.Distributed.Process.Management.Internal.Trace.Remote as Trace
153 ( remoteTable
193 ( remoteTable
155−import Control.Distributed.Process.Internal.Trace.Types
195+import Control.Distributed.Process.Management.Internal.Trace.Tracer
198+import Control.Distributed.Process.Management.Internal.Trace.Types
156 ( TraceArg(..)
199 ( TraceArg(..)
158 , traceEvent
200 , traceEvent
159 , traceLogFmt
201 , traceLogFmt
160 , enableTrace
202 , enableTrace
162−import Control.Distributed.Process.Internal.Trace.Tracer
166import Control.Distributed.Process.Serializable (Serializable)
204import Control.Distributed.Process.Serializable (Serializable)
167import Control.Distributed.Process.Internal.Messaging
205import Control.Distributed.Process.Internal.Messaging
168 ( sendBinary
206 ( sendBinary
170 , closeImplicitReconnections
207 , closeImplicitReconnections
171 , impliesDeathOf
208 , impliesDeathOf
173import Control.Distributed.Process.Internal.Primitives
210import Control.Distributed.Process.Internal.Primitives
174 ( register
211 ( register
176 , receiveWait
212 , receiveWait
178 , sendChan
214 , sendChan
179− , catch
215+ , unwrapMessage
181−import Control.Distributed.Process.Internal.Types (SendPort)
221+import Control.Distributed.Process.Internal.Types (SendPort, Tracer(..))
182import qualified Control.Distributed.Process.Internal.Closure.BuiltIn as BuiltIn (remoteTable)
222import qualified Control.Distributed.Process.Internal.Closure.BuiltIn as BuiltIn (remoteTable)
183import Control.Distributed.Process.Internal.WeakTQueue (TQueue, writeTQueue)
223import Control.Distributed.Process.Internal.WeakTQueue (TQueue, writeTQueue)
184import qualified Control.Distributed.Process.Internal.StrictContainerAccessors as DAC
224import qualified Control.Distributed.Process.Internal.StrictContainerAccessors as DAC
185 ( mapMaybe
225 ( mapMaybe
186 , mapDefault
226 , mapDefault
188−import System.Environment (getEnv)
228+import Control.Monad.Catch (try)
229+import GHC.IO (IO(..), unsafeUnmask)
230+import GHC.Base ( maskAsyncExceptions# )
189import Unsafe.Coerce
232import Unsafe.Coerce
235+-- Remove these definitions when the fix for
236+-- https://ghc.haskell.org/trac/ghc/ticket/10149
237+-- is included in all supported compilers:
238+block :: IO a -> IO a
239+block (IO io) = IO $ maskAsyncExceptions# io
240+unblock :: IO a -> IO a
241+unblock = unsafeUnmask
191--------------------------------------------------------------------------------
243--------------------------------------------------------------------------------
192-- Initialization --
244-- Initialization --
@@ -210,7 +262,7 @@ newLocalNode transport rtable = do
210createBareLocalNode :: NT.EndPoint -> RemoteTable -> IO LocalNode
262createBareLocalNode :: NT.EndPoint -> RemoteTable -> IO LocalNode
211createBareLocalNode endPoint rtable = do
263createBareLocalNode endPoint rtable = do
212 unq <- randomIO
264 unq <- randomIO
213− state <- newMVar LocalNodeState
265+ state <- newMVar $ LocalNodeValid $ ValidLocalNodeState
214 { _localProcesses = Map.empty
266 { _localProcesses = Map.empty
215 , _localPidCounter = firstNonReservedProcessId
267 , _localPidCounter = firstNonReservedProcessId
216 , _localPidUnique = unq
268 , _localPidUnique = unq
@@ -221,10 +273,10 @@ createBareLocalNode endPoint rtable = do
221 , localEndPoint = endPoint
273 , localEndPoint = endPoint
222 , localState = state
274 , localState = state
223 , localCtrlChan = ctrlChan
275 , localCtrlChan = ctrlChan
224− , localTracer = InactiveTracer
276+ , localEventBus = MxEventBusInitialising
225 , remoteTable = rtable
277 , remoteTable = rtable
227− tracedNode <- startTracing node
279+ tracedNode <- startMxAgent node
229 -- Once the NC terminates, the endpoint isn't much use,
281 -- Once the NC terminates, the endpoint isn't much use,
230 void $ forkIO $ Exception.finally (runNodeController tracedNode)
282 void $ forkIO $ Exception.finally (runNodeController tracedNode)
@@ -242,38 +294,27 @@ createBareLocalNode endPoint rtable = do
242 , ctrlMsgSignal = SigShutdown
294 , ctrlMsgSignal = SigShutdown
245−startTracing :: LocalNode -> IO LocalNode
297+startMxAgent :: LocalNode -> IO LocalNode
246−startTracing node =
298+startMxAgent node = do
247− -- TODO: use tryGetEnv once we drop support for GHC 7.2.x
249− (getEnv "DISTRIBUTED_PROCESS_TRACE_ENABLED" >> procTracer node)
250− (\(_ :: IOError) -> return node)
251− where procTracer :: LocalNode -> IO LocalNode
253− -- see note [tracer/forkProcess races]
299+ -- see note [tracer/forkProcess races]
300+ let fork = forkProcess node
254− mv <- MVar.newEmptyMVar
301+ mv <- MVar.newEmptyMVar
255− pid <- forkProcess n (traceController mv)
302+ pid <- fork $ mxAgentController fork mv
256− wqRef <- MVar.takeMVar mv
303+ (tracer', wqRef, mxNew') <- MVar.takeMVar mv
257− return n { localTracer = (ActiveTracer pid wqRef) }
304+ return node { localEventBus = (MxEventBus pid tracer' wqRef mxNew') }
259−startDefaultTracer :: LocalNode -> IO (Maybe ProcessId)
306+startDefaultTracer :: LocalNode -> IO ()
260−startDefaultTracer node =
307+startDefaultTracer node' = do
261− Exception.catch
308+ let t = localEventBus node'
262− (getEnv "DISTRIBUTED_PROCESS_TRACE_ENABLED" >> go node)
263− (\(_ :: IOError) -> return Nothing)
264− where go :: LocalNode -> IO (Maybe ProcessId)
266− registerTraceController node'
267− pid <- forkProcess node' defaultTracer
268− enableTrace (localTracer node') pid
271− registerTraceController :: LocalNode -> IO ()
272− registerTraceController n =
273− let t = localTracer n in
274− case t of
309+ case t of
275− InactiveTracer -> return ()
310+ MxEventBus _ (Tracer pid _) _ _ -> do
276− (ActiveTracer pid _) -> runProcess n $ register "trace.controller" pid
311+ runProcess node' $ register "trace.controller" pid
312+ pid' <- forkProcess node' defaultTracer
313+ enableTrace (localEventBus node') pid'
314+ runProcess node' $ register "tracer.initial" pid'
317+-- TODO: we need a better mechanism for defining and registering services
278-- | Start and register the service processes on a node
319-- | Start and register the service processes on a node
279startServiceProcesses :: LocalNode -> IO ()
320startServiceProcesses :: LocalNode -> IO ()
@@ -282,53 +323,71 @@ startServiceProcesses node = do
282 -- the default tracer first, even though it might @nsend@ to the logger
323 -- the default tracer first, even though it might @nsend@ to the logger
283 -- before /that/ process has started - this is a totally harmless race
324 -- before /that/ process has started - this is a totally harmless race
284 -- however, so we deliberably ignore it
325 -- however, so we deliberably ignore it
285− mPid <- startDefaultTracer node
326+ startDefaultTracer node
287− Nothing -> return ()
288− Just pid -> runProcess node $ register "tracer.initial" pid
290 logger <- forkProcess node loop
327 logger <- forkProcess node loop
328+ runProcess node $ do
329+ register "logger" logger
330+ -- The trace.logger is used for tracing to the console to avoid feedback
331+ -- loops during tracing if the user reregisters the "logger" with a custom
332+ -- process which uses 'send' or other primitives which are traced.
291− runProcess node $ register "logger" logger
333+ register "trace.logger" logger
293− loop = do
335+ loop = do
294− receiveWait
336+ receiveWait
295− [ match $ \((time, pid, string) ::(String, ProcessId, String)) -> do
337+ [ match $ \(SayMessage time pid string) -> do
338+ let time' = formatTime defaultTimeLocale "%c" time
296− liftIO . hPutStrLn stderr $ time ++ " " ++ show pid ++ ": " ++ string
339+ liftIO . hPutStrLn stderr $ time' ++ " " ++ show pid ++ ": " ++ string
298− , match $ \((time, string) :: (String, String)) -> do
341+ , match $ \((time, string) :: (String, String)) -> do
299− -- this is a 'trace' message from the local node tracer
342+ -- this is a 'trace' message from the local node tracer
300− liftIO . hPutStrLn stderr $ time ++ " [trace] " ++ string
343+ liftIO . hPutStrLn stderr $ time ++ " [trace] " ++ string
302− , match $ \(ch :: SendPort ()) -> -- a shutdown request
345+ , match $ \(ch :: SendPort ()) -> -- a shutdown request
303− sendChan ch ()
346+ sendChan ch ()
306−-- | Force-close a local node
349+-- | Force-close a local node, killing all processes on that node.
308−-- TODO: for now we just close the associated endpoint
309closeLocalNode :: LocalNode -> IO ()
350closeLocalNode :: LocalNode -> IO ()
310−closeLocalNode node =
351+closeLocalNode node = do
311− -- TODO: close all our processes, surely!?
352+ -- Kill processes after refilling the mvar. Otherwise, there is potential for
353+ -- deadlock as a dying process tries to get the mvar while masking exceptions
355+ join $ modifyMVar (localState node) $ \st -> case st of
356+ LocalNodeValid vst -> do
357+ return ( LocalNodeClosed
358+ , forM_ (vst ^. localProcesses) $ \lproc ->
359+ -- Semantics of 'throwTo' guarantee that target thread will get
360+ -- delivered an exception. Therefore, target thread will be
361+ -- killed eventually and that's as good as we can do. No need
362+ -- to wait for thread to actually finish dying.
363+ killThread (processThread lproc)
365+ LocalNodeClosed -> return (LocalNodeClosed, return ())
366+ -- This call will have the effect of shutting down the NC as well (see
367+ -- 'createBareLocalNode').
312 NT.closeEndPoint (localEndPoint node)
368 NT.closeEndPoint (localEndPoint node)
314-- | Run a process on a local node and wait for it to finish
370-- | Run a process on a local node and wait for it to finish
315runProcess :: LocalNode -> Process () -> IO ()
371runProcess :: LocalNode -> Process () -> IO ()
316runProcess node proc = do
372runProcess node proc = do
317 done <- newEmptyMVar
373 done <- newEmptyMVar
318− tid <- myThreadId
374+ -- TODO; When forkProcess inherits the masking state, protect the forked
375+ -- thread against async exceptions that could occur before 'try' is evaluated.
319− void $ forkProcess node $ do
376+ void $ forkProcess node $ try proc >>= liftIO . putMVar done
320− catch (proc `finally` liftIO (putMVar done ()))
377+ takeMVar done >>= either (throwIO :: SomeException -> IO a) return
321− (\(ex :: SomeException) -> liftIO $ throwTo tid ex)
324-- | Spawn a new process on a local node
379-- | Spawn a new process on a local node
325forkProcess :: LocalNode -> Process () -> IO ProcessId
380forkProcess :: LocalNode -> Process () -> IO ProcessId
381+forkProcess node proc = do
382+ ms <- getMaskingState
326−forkProcess node proc = modifyMVar (localState node) startProcess
383+ modifyMVarMasked (localState node) (startProcess ms)
385+ startProcess :: MaskingState
328− startProcess :: LocalNodeState -> IO (LocalNodeState, ProcessId)
387+ -> IO (LocalNodeState, ProcessId)
329− startProcess st = do
388+ startProcess ms (LocalNodeValid vst) = do
330− let lpid = LocalProcessId { lpidCounter = st ^. localPidCounter
389+ let lpid = LocalProcessId { lpidCounter = vst ^. localPidCounter
331− , lpidUnique = st ^. localPidUnique
390+ , lpidUnique = vst ^. localPidUnique
333 let pid = ProcessId { processNodeId = localNodeId node
392 let pid = ProcessId { processNodeId = localNodeId node
334 , processLocalId = lpid
393 , processLocalId = lpid
@@ -348,53 +407,78 @@ forkProcess node proc = modifyMVar (localState node) startProcess
348 , processThread = tid
407 , processThread = tid
349 , processNode = node
408 , processNode = node
410+ -- Rewrite this code when this is fixed:
411+ -- https://ghc.haskell.org/trac/ghc/ticket/10149
412+ let unmask = case ms of
414+ MaskedInterruptible -> block
415+ MaskedUninterruptible -> id
351− tid' <- forkIO $ do
416+ tid' <- uninterruptibleMask_ $ forkIO $ do
352− reason <- Exception.catch
417+ reason <- Exception.catches
353− (runLocalProcess lproc proc >> return DiedNormal)
418+ (unmask $ runLocalProcess lproc proc >> return DiedNormal)
419+ [ (Exception.Handler (\ex@(ProcessExitException from msg) -> do
420+ mMsg <- unwrapMessage msg :: IO (Maybe String)
422+ Nothing -> return $ DiedException $ show ex
423+ Just m -> return $ DiedException ("exit-from=" ++ (show from) ++ ",reason=" ++ m)))
424+ , (Exception.Handler
354− (return . DiedException . (show :: SomeException -> String))
425+ (return . DiedException . (show :: SomeException -> String)))]
356 -- [Unified: Table 4, rules termination and exiting]
427 -- [Unified: Table 4, rules termination and exiting]
357− modifyMVar_ (localState node) (cleanupProcess pid)
428+ mconns <- modifyValidLocalState node (cleanupProcess pid)
429+ -- XXX: Revisit after agreeing on the bigger picture for the semantics
430+ -- of transport operations.
431+ -- https://github.com/haskell-distributed/distributed-process/issues/204
432+ forM_ mconns $ forkIO . mapM_ NT.close
358 writeChan (localCtrlChan node) NCMsg
434 writeChan (localCtrlChan node) NCMsg
359 { ctrlMsgSender = ProcessIdentifier pid
435 { ctrlMsgSender = ProcessIdentifier pid
360 , ctrlMsgSignal = Died (ProcessIdentifier pid) reason
436 , ctrlMsgSignal = Died (ProcessIdentifier pid) reason
362 return (tid', lproc)
438 return (tid', lproc)
365 -- see note [tracer/forkProcess races]
440 -- see note [tracer/forkProcess races]
366− trace node (TraceEvSpawned pid)
441+ trace node (MxSpawned pid)
368 if lpidCounter lpid == maxBound
443 if lpidCounter lpid == maxBound
445+ -- TODO: this doesn't look right at all - how do we know
446+ -- that newUnique represents a process id that is available!?
370 newUnique <- randomIO
447 newUnique <- randomIO
448+ return ( LocalNodeValid
371− return ( (localProcessWithId lpid ^= Just lproc)
449+ $ (localProcessWithId lpid ^= Just lproc)
372 . (localPidCounter ^= firstNonReservedProcessId)
450 . (localPidCounter ^= firstNonReservedProcessId)
373 . (localPidUnique ^= newUnique)
451 . (localPidUnique ^= newUnique)
456+ return ( LocalNodeValid
378− return ( (localProcessWithId lpid ^= Just lproc)
457+ $ (localProcessWithId lpid ^= Just lproc)
379 . (localPidCounter ^: (+ 1))
458 . (localPidCounter ^: (+ 1))
462+ startProcess _ LocalNodeClosed =
463+ throwIO $ NodeClosedException $ localNodeId node
384− cleanupProcess :: ProcessId -> LocalNodeState -> IO LocalNodeState
465+ cleanupProcess :: ProcessId
466+ -> ValidLocalNodeState
467+ -> IO (ValidLocalNodeState, [NT.Connection])
385− cleanupProcess pid st = do
468+ cleanupProcess pid vst = do
386 let pid' = ProcessIdentifier pid
469 let pid' = ProcessIdentifier pid
387− let (affected, unaffected) = Map.partitionWithKey (\(fr, _to) !_v -> impliesDeathOf pid' fr) (st ^. localConnections)
470+ let (affected, unaffected) = Map.partitionWithKey (\(fr, _to) !_v -> impliesDeathOf pid' fr) (vst ^. localConnections)
388− mapM_ (NT.close . fst) (Map.elems affected)
389− return $ (localProcessWithId (processLocalId pid) ^= Nothing)
471+ return ( (localProcessWithId (processLocalId pid) ^= Nothing)
390− . (localConnections ^= unaffected)
472+ . (localConnections ^= unaffected)
474+ , map fst $ Map.elems affected
393-- note [tracer/forkProcess races]
477-- note [tracer/forkProcess races]
395-- Our startTracing function uses forkProcess to start the trace controller
479-- Our startTracing function uses forkProcess to start the trace controller
396-- process, and of course forkProcess attempts to call traceEvent once the
480-- process, and of course forkProcess attempts to call traceEvent once the
397−-- process has started. This is harmless, as the localTracer is not updated
481+-- process has started. This is harmless, as the localEventBus is not updated
398-- until /after/ the initial forkProcess completes, so the first call to
482-- until /after/ the initial forkProcess completes, so the first call to
399-- traceEvent behaves as if tracing were disabled (i.e., it is ignored).
483-- traceEvent behaves as if tracing were disabled (i.e., it is ignored).
@@ -408,7 +492,7 @@ type IncomingConnection = (NT.EndPointAddress, IncomingTarget)
408data IncomingTarget =
492data IncomingTarget =
410 | ToProc ProcessId (Weak (CQueue Message))
494 | ToProc ProcessId (Weak (CQueue Message))
411− | ToChan TypedChannel
495+ | ToChan SendPortId TypedChannel
414data ConnectionState = ConnectionState {
498data ConnectionState = ConnectionState {
@@ -435,6 +519,7 @@ incomingFrom addr = aux >>> DAC.mapDefault Set.empty addr
436handleIncomingMessages :: LocalNode -> IO ()
520handleIncomingMessages :: LocalNode -> IO ()
437handleIncomingMessages node = go initConnectionState
521handleIncomingMessages node = go initConnectionState
522+ `Exception.catch` \(NodeClosedException _) -> return ()
439 go :: ConnectionState -> IO ()
524 go :: ConnectionState -> IO ()
440 go !st = do
525 go !st = do
@@ -443,13 +528,14 @@ handleIncomingMessages node = go initConnectionState
443 NT.ConnectionOpened cid rel theirAddr ->
528 NT.ConnectionOpened cid rel theirAddr ->
444 if rel == NT.ReliableOrdered
529 if rel == NT.ReliableOrdered
446− trace node (TraceEvConnected cid)
531+ trace node (MxConnected cid theirAddr)
448 (incomingAt cid ^= Just (theirAddr, Uninit))
533 (incomingAt cid ^= Just (theirAddr, Uninit))
449 . (incomingFrom theirAddr ^: Set.insert cid)
534 . (incomingFrom theirAddr ^: Set.insert cid)
452− else invalidRequest cid st
537+ else invalidRequest cid st $
538+ "attempt to connect with unsupported reliability " ++ show rel
453 NT.Received cid payload ->
539 NT.Received cid payload ->
454 case st ^. incomingAt cid of
540 case st ^. incomingAt cid of
455 Just (_, ToProc pid weakQueue) -> do
541 Just (_, ToProc pid weakQueue) -> do
@@ -459,15 +545,19 @@ handleIncomingMessages node = go initConnectionState
459 -- it from the NC state? (and same for channels, below)
545 -- it from the NC state? (and same for channels, below)
460 let msg = payloadToMessage payload
546 let msg = payloadToMessage payload
461 enqueue queue msg -- 'enqueue' is strict
547 enqueue queue msg -- 'enqueue' is strict
462− trace node (TraceEvReceived pid msg)
548+ trace node (MxReceived pid msg)
464− Just (_, ToChan (TypedChannel chan')) -> do
550+ Just (_, ToChan chId (TypedChannel chan')) -> do
465 mChan <- deRefWeak chan'
551 mChan <- deRefWeak chan'
466 -- If mChan is Nothing, the process has given up the read end of
552 -- If mChan is Nothing, the process has given up the read end of
467 -- the channel and we simply ignore the incoming message
553 -- the channel and we simply ignore the incoming message
468− forM_ mChan $ \chan -> atomically $
554+ forM_ mChan $ \chan -> do
555+ msg' <- atomically $ do
556+ msg <- return $! decode (BSL.fromChunks payload)
469− -- We make sure the message is fully decoded when it is enqueued
557+ -- We make sure the message is fully decoded when it is enqueued
470− writeTQueue chan $! decode (BSL.fromChunks payload)
558+ writeTQueue chan msg
560+ trace node $ MxReceivedPort chId $ unsafeCreateUnencodedMessage msg'
472 Just (_, ToNode) -> do
562 Just (_, ToNode) -> do
473 let ctrlMsg = decode . BSL.fromChunks $ payload
563 let ctrlMsg = decode . BSL.fromChunks $ payload
@@ -477,38 +567,49 @@ handleIncomingMessages node = go initConnectionState
477 case decode (BSL.fromChunks payload) of
567 case decode (BSL.fromChunks payload) of
478 ProcessIdentifier pid -> do
568 ProcessIdentifier pid -> do
479 let lpid = processLocalId pid
569 let lpid = processLocalId pid
480− mProc <- withMVar state $ return . (^. localProcessWithId lpid)
570+ mProc <- withValidLocalState node $ return . (^. localProcessWithId lpid)
481 case mProc of
571 case mProc of
482 Just proc ->
572 Just proc ->
483 go (incomingAt cid ^= Just (src, ToProc pid (processWeakQ proc)) $ st)
573 go (incomingAt cid ^= Just (src, ToProc pid (processWeakQ proc)) $ st)
484 Nothing ->
574 Nothing ->
485− invalidRequest cid st
575+ -- incoming attempt to connect to unknown process - might
577+ go (incomingAt cid ^= Nothing $ st)
486 SendPortIdentifier chId -> do
578 SendPortIdentifier chId -> do
487 let lcid = sendPortLocalId chId
579 let lcid = sendPortLocalId chId
488 lpid = processLocalId (sendPortProcessId chId)
580 lpid = processLocalId (sendPortProcessId chId)
489− mProc <- withMVar state $ return . (^. localProcessWithId lpid)
581+ mProc <- withValidLocalState node $ return . (^. localProcessWithId lpid)
490 case mProc of
582 case mProc of
491 Just proc -> do
583 Just proc -> do
492 mChannel <- withMVar (processState proc) $ return . (^. typedChannelWithId lcid)
584 mChannel <- withMVar (processState proc) $ return . (^. typedChannelWithId lcid)
493 case mChannel of
585 case mChannel of
494 Just channel ->
586 Just channel ->
495− go (incomingAt cid ^= Just (src, ToChan channel) $ st)
587+ go (incomingAt cid ^= Just (src, ToChan chId channel) $ st)
496 Nothing ->
588 Nothing ->
497− invalidRequest cid st
589+ invalidRequest cid st $
590+ "incoming attempt to connect to unknown channel of"
591+ ++ " process " ++ show (sendPortProcessId chId)
498 Nothing ->
592 Nothing ->
499− invalidRequest cid st
593+ -- incoming attempt to connect to channel of unknown
594+ -- process - might be dead already
595+ go (incomingAt cid ^= Nothing $ st)
500 NodeIdentifier nid ->
596 NodeIdentifier nid ->
501 if nid == localNodeId node
597 if nid == localNodeId node
502 then go (incomingAt cid ^= Just (src, ToNode) $ st)
598 then go (incomingAt cid ^= Just (src, ToNode) $ st)
503− else invalidRequest cid st
599+ else invalidRequest cid st $
600+ "incoming attempt to connect to a different node -"
601+ ++ " I'm " ++ show (localNodeId node)
602+ ++ " but the remote peer wants to connect to "
504 Nothing ->
604 Nothing ->
505 invalidRequest cid st
605 invalidRequest cid st
606+ "message received from an unknown connection"
506 NT.ConnectionClosed cid ->
607 NT.ConnectionClosed cid ->
507 case st ^. incomingAt cid of
608 case st ^. incomingAt cid of
508 Nothing ->
609 Nothing ->
509− invalidRequest cid st
610+ invalidRequest cid st "closed unknown connection"
510 Just (src, _) -> do
611 Just (src, _) -> do
511− trace node (TraceEvDisconnected cid)
612+ trace node (MxDisconnected cid src)
512 go ( (incomingAt cid ^= Nothing)
613 go ( (incomingAt cid ^= Nothing)
513 . (incomingFrom src ^: Set.delete cid)
614 . (incomingFrom src ^: Set.delete cid)
@@ -537,19 +638,20 @@ handleIncomingMessages node = go initConnectionState
537 -- and we just give up
638 -- and we just give up
538 fail "Cloud Haskell fatal error: received unexpected multicast"
639 fail "Cloud Haskell fatal error: received unexpected multicast"
540− invalidRequest :: NT.ConnectionId -> ConnectionState -> IO ()
641+ invalidRequest :: NT.ConnectionId -> ConnectionState -> String -> IO ()
541− invalidRequest cid st = do
642+ invalidRequest cid st msg = do
542 -- TODO: We should treat this as a fatal error on the part of the remote
643 -- TODO: We should treat this as a fatal error on the part of the remote
543 -- node. That is, we should report the remote node as having died, and we
644 -- node. That is, we should report the remote node as having died, and we
544 -- should close incoming connections (this requires a Transport layer
645 -- should close incoming connections (this requires a Transport layer
545 -- extension).
646 -- extension).
546− traceEventFmtIO node "" [(TraceStr " [network] invalid request: "),
647+ traceEventFmtIO node "" [ TraceStr $ " [network] invalid request"
648+ ++ " (" ++ msg ++ "): "
547− (Trace cid)]
649+ , (Trace cid)
548 go ( incomingAt cid ^= Nothing
651 go ( incomingAt cid ^= Nothing
552− state = localState node
553 endpoint = localEndPoint node
655 endpoint = localEndPoint node
554 ctrlChan = localCtrlChan node
656 ctrlChan = localCtrlChan node
@@ -558,8 +660,9 @@ handleIncomingMessages node = go initConnectionState
558--------------------------------------------------------------------------------
660--------------------------------------------------------------------------------
560runNodeController :: LocalNode -> IO ()
662runNodeController :: LocalNode -> IO ()
561−runNodeController =
663+runNodeController node =
562− runReaderT (evalStateT (unNC nodeController) initNCState)
664+ runReaderT (evalStateT (unNC nodeController) initNCState) node
665+ `Exception.catch` \(NodeClosedException _) -> return ()
564--------------------------------------------------------------------------------
667--------------------------------------------------------------------------------
565-- Internal data types --
668-- Internal data types --
@@ -567,20 +670,26 @@ runNodeController =
568data NCState = NCState
671data NCState = NCState
569 { -- Mapping from remote processes to linked local processes
672 { -- Mapping from remote processes to linked local processes
570− _links :: !(Map Identifier (Set ProcessId))
673+ _links :: !(BiMultiMap Identifier ProcessId ())
571 -- Mapping from remote processes to monitoring local processes
674 -- Mapping from remote processes to monitoring local processes
572− , _monitors :: !(Map Identifier (Set (ProcessId, MonitorRef)))
675+ , _monitors :: !(BiMultiMap Identifier ProcessId MonitorRef)
573 -- Process registry: names and where they live, mapped to the PIDs
676 -- Process registry: names and where they live, mapped to the PIDs
574 , _registeredHere :: !(Map String ProcessId)
677 , _registeredHere :: !(Map String ProcessId)
575 , _registeredOnNodes :: !(Map ProcessId [(NodeId,Int)])
678 , _registeredOnNodes :: !(Map ProcessId [(NodeId,Int)])
578newtype NC a = NC { unNC :: StateT NCState (ReaderT LocalNode IO) a }
681newtype NC a = NC { unNC :: StateT NCState (ReaderT LocalNode IO) a }
682+ deriving ( Applicative
686+ , MonadState NCState
579− deriving (Functor, Monad, MonadIO, MonadState NCState, MonadReader LocalNode)
687+ , MonadReader LocalNode
581initNCState :: NCState
690initNCState :: NCState
582−initNCState = NCState { _links = Map.empty
691+initNCState = NCState { _links = BiMultiMap.empty
583− , _monitors = Map.empty
692+ , _monitors = BiMultiMap.empty
584 , _registeredHere = Map.empty
693 , _registeredHere = Map.empty
585 , _registeredOnNodes = Map.empty
694 , _registeredOnNodes = Map.empty
@@ -596,19 +705,46 @@ instance Show ProcessKillException where
596 show (ProcessKillException pid reason) =
705 show (ProcessKillException pid reason) =
597 "killed-by=" ++ show pid ++ ",reason=" ++ reason
706 "killed-by=" ++ show pid ++ ",reason=" ++ reason
708+ncSendToProcess :: ProcessId -> Message -> NC ()
709+ncSendToProcess = ncSendToProcessAndTrace True
711+ncSendToProcessAndTrace :: Bool -> ProcessId -> Message -> NC ()
712+ncSendToProcessAndTrace shouldTrace pid msg = do
714+ if processNodeId pid == localNodeId node
715+ then ncEffectLocalSendAndTrace shouldTrace node pid msg
716+ else liftIO $ sendBinary node
717+ (NodeIdentifier $ localNodeId node)
718+ (NodeIdentifier $ processNodeId pid)
719+ WithImplicitReconnect
720+ NCMsg { ctrlMsgSender = NodeIdentifier $ localNodeId node
721+ , ctrlMsgSignal = UnreliableSend (processLocalId pid) msg
724+ncSendToNode :: NodeId -> NCMsg -> NC ()
725+ncSendToNode to msg = do
727+ liftIO $ if to == localNodeId node
728+ then writeChan (localCtrlChan node) $! msg
729+ else sendBinary node
730+ (NodeIdentifier $ localNodeId node)
732+ WithImplicitReconnect
599--------------------------------------------------------------------------------
735--------------------------------------------------------------------------------
600-- Tracing/Debugging --
736-- Tracing/Debugging --
601--------------------------------------------------------------------------------
737--------------------------------------------------------------------------------
603−-- [Issue #104]
739+-- [Issue #104 / DP-13]
605traceNotifyDied :: LocalNode -> Identifier -> DiedReason -> NC ()
741traceNotifyDied :: LocalNode -> Identifier -> DiedReason -> NC ()
606traceNotifyDied node ident reason =
742traceNotifyDied node ident reason =
607 -- TODO: sendPortDied notifications
743 -- TODO: sendPortDied notifications
608 liftIO $ withLocalTracer node $ \t ->
744 liftIO $ withLocalTracer node $ \t ->
609 case ident of
745 case ident of
610− (NodeIdentifier nid) -> traceEvent t (TraceEvNodeDied nid reason)
746+ (NodeIdentifier nid) -> traceEvent t (MxNodeDied nid reason)
611− (ProcessIdentifier pid) -> traceEvent t (TraceEvDied pid reason)
747+ (ProcessIdentifier pid) -> traceEvent t (MxProcessDied pid reason)
612 _ -> return ()
748 _ -> return ()
614traceEventFmtIO :: LocalNode
750traceEventFmtIO :: LocalNode
@@ -618,11 +754,11 @@ traceEventFmtIO :: LocalNode
618traceEventFmtIO node fmt args =
754traceEventFmtIO node fmt args =
619 withLocalTracer node $ \t -> traceLogFmt t fmt args
755 withLocalTracer node $ \t -> traceLogFmt t fmt args
621−trace :: LocalNode -> TraceEvent -> IO ()
757+trace :: LocalNode -> MxEvent -> IO ()
622trace node ev = withLocalTracer node $ \t -> traceEvent t ev
758trace node ev = withLocalTracer node $ \t -> traceEvent t ev
624−withLocalTracer :: LocalNode -> (Tracer -> IO ()) -> IO ()
760+withLocalTracer :: LocalNode -> (MxEventBus -> IO ()) -> IO ()
625−withLocalTracer node act = act (localTracer node)
761+withLocalTracer node act = act (localEventBus node)
627--------------------------------------------------------------------------------
763--------------------------------------------------------------------------------
628-- Core functionality --
764-- Core functionality --
@@ -638,11 +774,7 @@ nodeController = do
638 -- [Unified: Table 7, rule nc_forward]
774 -- [Unified: Table 7, rule nc_forward]
639 case destNid (ctrlMsgSignal msg) of
775 case destNid (ctrlMsgSignal msg) of
640 Just nid' | nid' /= localNodeId node ->
776 Just nid' | nid' /= localNodeId node ->
641− liftIO $ sendBinary node
777+ ncSendToNode nid' msg
643− (NodeIdentifier nid')
644− WithImplicitReconnect
647 return ()
779 return ()
@@ -665,6 +797,8 @@ nodeController = do
665 ncEffectWhereIs from label
797 ncEffectWhereIs from label
666 NCMsg _ (NamedSend label msg') ->
798 NCMsg _ (NamedSend label msg') ->
667 ncEffectNamedSend label msg'
799 ncEffectNamedSend label msg'
800+ NCMsg _ (UnreliableSend lpid msg') ->
801+ ncEffectLocalSend node (ProcessId (localNodeId node) lpid) msg'
668 NCMsg _ (LocalSend to msg') ->
802 NCMsg _ (LocalSend to msg') ->
669 ncEffectLocalSend node to msg'
803 ncEffectLocalSend node to msg'
670 NCMsg _ (LocalPortSend to msg') ->
804 NCMsg _ (LocalPortSend to msg') ->
@@ -678,8 +812,9 @@ nodeController = do
678 NCMsg _ SigShutdown ->
812 NCMsg _ SigShutdown ->
679 liftIO $ do
813 liftIO $ do
680 NT.closeEndPoint (localEndPoint node)
814 NT.closeEndPoint (localEndPoint node)
681− `Exception.finally` throwIO ThreadKilled
815+ `Exception.finally` throwIO (NodeClosedException $ localNodeId node)
682− -- ThreadKilled seems to make more sense than fail/error here
816+ NCMsg (ProcessIdentifier from) (GetNodeStats nid) ->
817+ ncEffectGetNodeStats from nid
683 unexpected ->
818 unexpected ->
684 error $ "nodeController: unexpected message " ++ show unexpected
819 error $ "nodeController: unexpected message " ++ show unexpected
@@ -697,22 +832,18 @@ ncEffectMonitor from them mRef = do
697 case (shouldLink, isLocal node (ProcessIdentifier from)) of
832 case (shouldLink, isLocal node (ProcessIdentifier from)) of
698 (True, _) -> -- [Unified: first rule]
833 (True, _) -> -- [Unified: first rule]
699 case mRef of
834 case mRef of
700− Just ref -> modify' $ monitorsFor them ^: Set.insert (from, ref)
835+ Just ref -> modify' $ monitors ^: BiMultiMap.insert them from ref
701− Nothing -> modify' $ linksFor them ^: Set.insert from
836+ Nothing -> modify' $ links ^: BiMultiMap.insert them from ()
702 (False, True) -> -- [Unified: second rule]
837 (False, True) -> -- [Unified: second rule]
703 notifyDied from them DiedUnknownId mRef
838 notifyDied from them DiedUnknownId mRef
704 (False, False) -> -- [Unified: third rule]
839 (False, False) -> -- [Unified: third rule]
705 -- TODO: this is the right sender according to the Unified semantics,
840 -- TODO: this is the right sender according to the Unified semantics,
706 -- but perhaps having 'them' as the sender would make more sense
841 -- but perhaps having 'them' as the sender would make more sense
707 -- (see also: notifyDied)
842 -- (see also: notifyDied)
708− liftIO $ sendBinary node
709− (NodeIdentifier $ localNodeId node)
710− (NodeIdentifier $ processNodeId from)
843+ ncSendToNode (processNodeId from) $ NCMsg
711− WithImplicitReconnect
713− { ctrlMsgSender = NodeIdentifier (localNodeId node)
844+ { ctrlMsgSender = NodeIdentifier (localNodeId node)
714− , ctrlMsgSignal = Died them DiedUnknownId
845+ , ctrlMsgSignal = Died them DiedUnknownId
717-- [Unified: Table 11]
848-- [Unified: Table 11]
718ncEffectUnlink :: ProcessId -> Identifier -> NC ()
849ncEffectUnlink :: ProcessId -> Identifier -> NC ()
@@ -726,7 +857,7 @@ ncEffectUnlink from them = do
726 postAsMessage from $ DidUnlinkNode nid
857 postAsMessage from $ DidUnlinkNode nid
727 SendPortIdentifier cid ->
858 SendPortIdentifier cid ->
728 postAsMessage from $ DidUnlinkPort cid
859 postAsMessage from $ DidUnlinkPort cid
729− modify' $ linksFor them ^: Set.delete from
860+ modify' $ links ^: BiMultiMap.delete them from ()
731-- [Unified: Table 11]
862-- [Unified: Table 11]
732ncEffectUnmonitor :: ProcessId -> MonitorRef -> NC ()
863ncEffectUnmonitor :: ProcessId -> MonitorRef -> NC ()
@@ -734,7 +865,7 @@ ncEffectUnmonitor from ref = do
734 node <- ask
865 node <- ask
735 when (isLocal node (ProcessIdentifier from)) $
866 when (isLocal node (ProcessIdentifier from)) $
736 postAsMessage from $ DidUnmonitor ref
867 postAsMessage from $ DidUnmonitor ref
737− modify' $ monitorsFor (monitorRefIdent ref) ^: Set.delete (from, ref)
868+ modify' $ monitors ^: BiMultiMap.delete (monitorRefIdent ref) from ref
739-- [Unified: Table 12]
870-- [Unified: Table 12]
740ncEffectDied :: Identifier -> DiedReason -> NC ()
871ncEffectDied :: Identifier -> DiedReason -> NC ()
@@ -749,7 +880,7 @@ ncEffectDied ident reason = do
749 let localOnly = case ident of NodeIdentifier _ -> True ; _ -> False
880 let localOnly = case ident of NodeIdentifier _ -> True ; _ -> False
751 forM_ (Map.toList affectedLinks) $ \(them, uss) ->
882 forM_ (Map.toList affectedLinks) $ \(them, uss) ->
752− forM_ uss $ \us ->
883+ forM_ uss $ \(us, _) ->
753 when (localOnly <= isLocal node (ProcessIdentifier us)) $
884 when (localOnly <= isLocal node (ProcessIdentifier us)) $
754 notifyDied us them reason Nothing
885 notifyDied us them reason Nothing
@@ -758,9 +889,35 @@ ncEffectDied ident reason = do
758 when (localOnly <= isLocal node (ProcessIdentifier us)) $
889 when (localOnly <= isLocal node (ProcessIdentifier us)) $
759 notifyDied us them reason (Just ref)
890 notifyDied us them reason (Just ref)
892+ -- Notify remote nodes that the process died so it can be removed from monitor
894+ mapM_ (forwardDeath node) $
895+ [ nid | ProcessIdentifier pid <- [ident]
896+ , i <- Set.toList $ Set.union
897+ (Set.map fst $ BiMultiMap.lookupBy2nd pid unaffectedLinks)
898+ (Set.map fst $ BiMultiMap.lookupBy2nd pid unaffectedMons)
899+ , let nid = nodeOf i
900+ , nid /= localNodeId node
903+ -- Delete monitors in the local node.
904+ let deleteDeads :: (Ord a, Ord v)
905+ => BiMultiMap a ProcessId v -> BiMultiMap a ProcessId v
906+ deleteDeads = case ident of
907+ -- deleteAllBy2nd is faster than partitionWithKeyBy2nd
908+ ProcessIdentifier pid -> BiMultiMap.deleteAllBy2nd pid
909+ _ -> snd . BiMultiMap.partitionWithKeyBy2nd
910+ (\pid _ -> ident `impliesDeathOf` ProcessIdentifier pid)
911+ unaffectedLinks' = deleteDeads unaffectedLinks
912+ unaffectedMons' = deleteDeads unaffectedMons
761− modify' $ (links ^= unaffectedLinks) . (monitors ^= unaffectedMons)
914+ modify' $ (links ^= unaffectedLinks') . (monitors ^= unaffectedMons')
916+ -- we now consider all labels for this identifier unregistered
763− modify' $ registeredHere ^: Map.filter (\pid -> not $ ident `impliesDeathOf` ProcessIdentifier pid)
917+ let toDrop pid = not $ ident `impliesDeathOf` ProcessIdentifier pid
918+ (keepNames, dropNames) <- Map.partition toDrop <$> gets (^. registeredHere)
919+ mapM_ (\(p, l) -> liftIO $ trace node (MxUnRegistered l p)) (Map.toList dropNames)
920+ modify' $ registeredHere ^= keepNames
765 remaining <- fmap Map.toList (gets (^. registeredOnNodes)) >>=
922 remaining <- fmap Map.toList (gets (^. registeredOnNodes)) >>=
766 mapM (\(pid,nidlist) ->
923 mapM (\(pid,nidlist) ->
@@ -768,20 +925,15 @@ ncEffectDied ident reason = do
769 do forM_ nidlist $ \(nid,_) ->
926 do forM_ nidlist $ \(nid,_) ->
770 when (not $ isLocal node (NodeIdentifier nid))
927 when (not $ isLocal node (NodeIdentifier nid))
771− (forwardNameDeath node nid)
928+ (forwardDeath node nid)
772 return Nothing
929 return Nothing
773 False -> return $ Just (pid,nidlist) )
930 False -> return $ Just (pid,nidlist) )
774 modify' $ registeredOnNodes ^= (Map.fromList (catMaybes remaining))
931 modify' $ registeredOnNodes ^= (Map.fromList (catMaybes remaining))
776− forwardNameDeath node nid =
933+ forwardDeath node nid = ncSendToNode nid
777− liftIO $ sendBinary node
778− (NodeIdentifier $ localNodeId node)
779− (NodeIdentifier $ nid)
780− WithImplicitReconnect
782− { ctrlMsgSender = NodeIdentifier (localNodeId node)
934+ NCMsg { ctrlMsgSender = NodeIdentifier (localNodeId node)
783− , ctrlMsgSignal = Died ident reason
935+ , ctrlMsgSignal = Died ident reason
786-- [Unified: Table 13]
938-- [Unified: Table 13]
787ncEffectSpawn :: ProcessId -> Closure (Process ()) -> SpawnRef -> NC ()
939ncEffectSpawn :: ProcessId -> Closure (Process ()) -> SpawnRef -> NC ()
@@ -789,17 +941,13 @@ ncEffectSpawn pid cProc ref = do
789 mProc <- unClosure cProc
941 mProc <- unClosure cProc
790 -- If the closure does not exist, we spawn a process that throws an exception
942 -- If the closure does not exist, we spawn a process that throws an exception
791 -- This allows the remote node to find out what's happening
943 -- This allows the remote node to find out what's happening
792− -- TODO:
944+ -- TODO:
793 let proc = case mProc of
945 let proc = case mProc of
794 Left err -> fail $ "Error: Could not resolve closure: " ++ err
946 Left err -> fail $ "Error: Could not resolve closure: " ++ err
795 Right p -> p
947 Right p -> p
796 node <- ask
948 node <- ask
797 pid' <- liftIO $ forkProcess node proc
949 pid' <- liftIO $ forkProcess node proc
798− liftIO $ sendMessage node
950+ ncSendToProcess pid $ unsafeCreateUnencodedMessage $ DidSpawn ref pid'
799− (NodeIdentifier (localNodeId node))
800− (ProcessIdentifier pid)
801− WithImplicitReconnect
804-- Unified semantics does not explicitly describe how to implement 'register',
952-- Unified semantics does not explicitly describe how to implement 'register',
805-- but mentions it's "very similar to nsend" (Table 14)
953-- but mentions it's "very similar to nsend" (Table 14)
@@ -817,17 +965,19 @@ ncEffectRegister from label atnode mPid reregistration = do
817 return $ (isNothing currentVal /= reregistration) &&
965 return $ (isNothing currentVal /= reregistration) &&
818 (not (isLocal node (ProcessIdentifier thepid) ) || isvalidlocal )
966 (not (isLocal node (ProcessIdentifier thepid) ) || isvalidlocal )
819 if isLocal node (NodeIdentifier atnode)
967 if isLocal node (NodeIdentifier atnode)
820− then do when (isOk) $
968+ then do when isOk $
821 do modify' $ registeredHereFor label ^= mPid
969 do modify' $ registeredHereFor label ^= mPid
822 updateRemote node currentVal mPid
970 updateRemote node currentVal mPid
823 case mPid of
971 case mPid of
824− (Just p) -> liftIO $ trace node (TraceEvRegistered p label)
972+ (Just p) -> do
974+ then liftIO $ trace node (MxUnRegistered (fromJust currentVal) label)
976+ liftIO $ trace node (MxRegistered p label)
825− Nothing -> liftIO $ trace node (TraceEvUnRegistered (fromJust currentVal) label)
977+ Nothing -> liftIO $ trace node (MxUnRegistered (fromJust currentVal) label)
826− liftIO $ sendMessage node
978+ newVal <- gets (^. registeredHereFor label)
827− (NodeIdentifier (localNodeId node))
979+ ncSendToProcess from $ unsafeCreateUnencodedMessage $
828− (ProcessIdentifier from)
829− WithImplicitReconnect
830− (RegisterReply label isOk)
980+ RegisterReply label isOk newVal
831 else let operation =
981 else let operation =
832 case reregistration of
982 case reregistration of
833 True -> flip decList
983 True -> flip decList
@@ -857,40 +1007,36 @@ ncEffectRegister from label atnode mPid reregistration = do
857 decList (x:xs) tag = x:decList xs tag
1007 decList (x:xs) tag = x:decList xs tag
858 forward node to reg =
1008 forward node to reg =
859 when (not $ isLocal node (NodeIdentifier to)) $
1009 when (not $ isLocal node (NodeIdentifier to)) $
860− liftIO $ sendBinary node
861− (ProcessIdentifier from)
863− WithImplicitReconnect
865− { ctrlMsgSender = ProcessIdentifier from
1010+ ncSendToNode to $ NCMsg { ctrlMsgSender = ProcessIdentifier from
866− , ctrlMsgSignal = reg
1011+ , ctrlMsgSignal = reg
870-- Unified semantics does not explicitly describe 'whereis'
1015-- Unified semantics does not explicitly describe 'whereis'
871ncEffectWhereIs :: ProcessId -> String -> NC ()
1016ncEffectWhereIs :: ProcessId -> String -> NC ()
872ncEffectWhereIs from label = do
1017ncEffectWhereIs from label = do
874 mPid <- gets (^. registeredHereFor label)
1018 mPid <- gets (^. registeredHereFor label)
875− liftIO $ sendMessage node
876− (NodeIdentifier (localNodeId node))
877− (ProcessIdentifier from)
878− WithImplicitReconnect
879− (WhereIsReply label mPid)
1019+ ncSendToProcess from $ unsafeCreateUnencodedMessage $ WhereIsReply label mPid
881-- [Unified: Table 14]
1021-- [Unified: Table 14]
882ncEffectNamedSend :: String -> Message -> NC ()
1022ncEffectNamedSend :: String -> Message -> NC ()
883ncEffectNamedSend label msg = do
1023ncEffectNamedSend label msg = do
884 mPid <- gets (^. registeredHereFor label)
1024 mPid <- gets (^. registeredHereFor label)
885 -- If mPid is Nothing, we just ignore the named send (as per Table 14)
1025 -- If mPid is Nothing, we just ignore the named send (as per Table 14)
886− forM_ mPid $ \pid -> postMessage pid msg
1026+ forM_ mPid $ \to ->
1027+ -- If this is a trace message we don't trace it to avoid entering a loop
1028+ -- where trace messages produce more trace messages.
1029+ ncSendToProcessAndTrace (label /= "trace.logger") to msg
888-- [Issue #DP-20]
1031-- [Issue #DP-20]
889ncEffectLocalSend :: LocalNode -> ProcessId -> Message -> NC ()
1032ncEffectLocalSend :: LocalNode -> ProcessId -> Message -> NC ()
890−ncEffectLocalSend node to msg =
1033+ncEffectLocalSend = ncEffectLocalSendAndTrace True
1035+ncEffectLocalSendAndTrace :: Bool -> LocalNode -> ProcessId -> Message -> NC ()
1036+ncEffectLocalSendAndTrace shouldTrace node to msg =
891 liftIO $ withLocalProc node to $ \p -> do
1037 liftIO $ withLocalProc node to $ \p -> do
892 enqueue (processQueue p) msg
1038 enqueue (processQueue p) msg
893− trace node (TraceEvReceived to msg)
1039+ when shouldTrace $ trace node (MxReceived to msg)
895-- [Issue #DP-20]
1041-- [Issue #DP-20]
896ncEffectLocalPortSend :: SendPortId -> Message -> NC ()
1042ncEffectLocalPortSend :: SendPortId -> Message -> NC ()
@@ -908,11 +1054,12 @@ ncEffectLocalPortSend from msg = do
908 -- If ch is Nothing, the process has given up the read end of
1054 -- If ch is Nothing, the process has given up the read end of
909 -- the channel and we simply ignore the incoming message - this
1055 -- the channel and we simply ignore the incoming message - this
910 ch <- deRefWeak chan'
1056 ch <- deRefWeak chan'
911− forM_ ch $ \chan -> deliverChan msg chan
1057+ forM_ ch $ \chan -> deliverChan node from msg chan
912− where deliverChan :: forall a . Message -> TQueue a -> IO ()
1058+ where deliverChan :: forall a . LocalNode -> SendPortId -> Message -> TQueue a -> IO ()
913− deliverChan (UnencodedMessage _ raw) chan' =
1059+ deliverChan n p (UnencodedMessage _ raw) chan' = do
914 atomically $ writeTQueue chan' ((unsafeCoerce raw) :: a)
1060 atomically $ writeTQueue chan' ((unsafeCoerce raw) :: a)
1061+ trace n (MxReceivedPort p $ unsafeCreateUnencodedMessage raw)
915− deliverChan (EncodedMessage _ _) _ =
1062+ deliverChan _ _ (EncodedMessage _ _) _ =
916 -- this will not happen unless someone screws with Primitives.hs
1063 -- this will not happen unless someone screws with Primitives.hs
917 error "invalid local channel delivery"
1064 error "invalid local channel delivery"
@@ -937,46 +1084,56 @@ ncEffectGetInfo from pid =
937 them = (ProcessIdentifier pid)
1084 them = (ProcessIdentifier pid)
939 node <- ask
1086 node <- ask
940− mProc <- liftIO $
1087+ mProc <- liftIO $ withValidLocalState node
941− withMVar (localState node) $ return . (^. localProcessWithId lpid)
1088+ $ return . (^. localProcessWithId lpid)
942 case mProc of
1089 case mProc of
943 Nothing -> dispatch (isLocal node (ProcessIdentifier from))
1090 Nothing -> dispatch (isLocal node (ProcessIdentifier from))
944− from node (ProcessInfoNone DiedUnknownId)
1091+ from (ProcessInfoNone DiedUnknownId)
945− Just _ -> do
1092+ Just proc -> do
946− itsLinks <- gets (^. linksFor them)
1093+ itsLinks <- Set.map fst . BiMultiMap.lookupBy1st them <$>
947− itsMons <- gets (^. monitorsFor them)
1094+ gets (^. links)
1095+ itsMons <- BiMultiMap.lookupBy1st them <$> gets (^. monitors)
948 registered <- gets (^. registeredHere)
1096 registered <- gets (^. registeredHere)
1097+ size <- liftIO $ queueSize $ processQueue $ proc
950 let reg = registeredNames registered
1099 let reg = registeredNames registered
951 dispatch (isLocal node (ProcessIdentifier from))
1100 dispatch (isLocal node (ProcessIdentifier from))
954 ProcessInfo {
1102 ProcessInfo {
955 infoNode = (processNodeId pid)
1103 infoNode = (processNodeId pid)
956 , infoRegisteredNames = reg
1104 , infoRegisteredNames = reg
957− -- we cannot populate this field
958− , infoMessageQueueLength = Nothing
1105+ , infoMessageQueueLength = size
959− , infoMonitors = Set.toList itsMons
1106+ , infoMonitors = Set.toList itsMons
960− , infoLinks = Set.toList itsLinks
1107+ , infoLinks = Set.toList itsLinks
962− where dispatch :: (Serializable a, Show a)
1109+ where dispatch :: (Serializable a)
964 -> ProcessId
1111 -> ProcessId
967 -> NC ()
1113 -> NC ()
968− dispatch True dest _ pInfo = postAsMessage dest $ pInfo
1114+ dispatch True dest pInfo = postAsMessage dest $ pInfo
969− dispatch False dest node pInfo = do
1115+ dispatch False dest pInfo =
970− liftIO $ sendMessage node
1116+ ncSendToProcess dest $ unsafeCreateUnencodedMessage pInfo
971− (NodeIdentifier (localNodeId node))
972− (ProcessIdentifier dest)
973− WithImplicitReconnect
976 registeredNames = Map.foldlWithKey (\ks k v -> if v == pid
1118 registeredNames = Map.foldlWithKey (\ks k v -> if v == pid
977 then (k:ks)
1119 then (k:ks)
978 else ks) []
1120 else ks) []
1122+ncEffectGetNodeStats :: ProcessId -> NodeId -> NC ()
1123+ncEffectGetNodeStats from _nid = do
1125+ ncState <- StateT.get
1126+ nodeState <- liftIO $ withValidLocalState node return
1129+ nodeStatsNode = localNodeId node
1130+ , nodeStatsRegisteredNames = Map.size $ ncState ^. registeredHere
1131+ , nodeStatsMonitors = BiMultiMap.size $ ncState ^. monitors
1132+ , nodeStatsLinks = BiMultiMap.size $ ncState ^. links
1133+ , nodeStatsProcesses = Map.size (nodeState ^. localProcesses)
1135+ postAsMessage from stats
980--------------------------------------------------------------------------------
1137--------------------------------------------------------------------------------
981-- Auxiliary --
1138-- Auxiliary --
982--------------------------------------------------------------------------------
1139--------------------------------------------------------------------------------
@@ -1003,14 +1160,10 @@ notifyDied dest src reason mRef = do
1003 throwException dest $ PortLinkException pid reason
1160 throwException dest $ PortLinkException pid reason
1004 (False, _, _) ->
1161 (False, _, _) ->
1005 -- The change in sender comes from [Unified: Table 10]
1162 -- The change in sender comes from [Unified: Table 10]
1006− liftIO $ sendBinary node
1007− (NodeIdentifier $ localNodeId node)
1008− (NodeIdentifier $ processNodeId dest)
1163+ ncSendToNode (processNodeId dest) $ NCMsg
1009− WithImplicitReconnect
1011− { ctrlMsgSender = NodeIdentifier (localNodeId node)
1164+ { ctrlMsgSender = NodeIdentifier (localNodeId node)
1012− , ctrlMsgSignal = Died src reason
1165+ , ctrlMsgSignal = Died src reason
1015-- | [Unified: Table 8]
1168-- | [Unified: Table 8]
1016destNid :: ProcessSignal -> Maybe NodeId
1169destNid :: ProcessSignal -> Maybe NodeId
@@ -1022,6 +1175,7 @@ destNid (Spawn _ _) = Nothing
1022destNid (Register _ _ _ _) = Nothing
1175destNid (Register _ _ _ _) = Nothing
1023destNid (WhereIs _) = Nothing
1176destNid (WhereIs _) = Nothing
1024destNid (NamedSend _ _) = Nothing
1177destNid (NamedSend _ _) = Nothing
1178+destNid (UnreliableSend _ _) = Nothing
1025-- We don't need to forward 'Died' signals; if monitoring/linking is setup,
1179-- We don't need to forward 'Died' signals; if monitoring/linking is setup,
1026-- then when a local process dies the monitoring/linking machinery will take
1180-- then when a local process dies the monitoring/linking machinery will take
1027-- care of notifying remote nodes
1181-- care of notifying remote nodes
@@ -1029,6 +1183,7 @@ destNid (Died _ _) = Nothing
1029destNid (Kill pid _) = Just $ processNodeId pid
1183destNid (Kill pid _) = Just $ processNodeId pid
1030destNid (Exit pid _) = Just $ processNodeId pid
1184destNid (Exit pid _) = Just $ processNodeId pid
1031destNid (GetInfo pid) = Just $ processNodeId pid
1185destNid (GetInfo pid) = Just $ processNodeId pid
1186+destNid (GetNodeStats nid) = Just nid
1032destNid (LocalSend pid _) = Just $ processNodeId pid
1187destNid (LocalSend pid _) = Just $ processNodeId pid
1033destNid (LocalPortSend cid _) = Just $ processNodeId (sendPortProcessId cid)
1188destNid (LocalPortSend cid _) = Just $ processNodeId (sendPortProcessId cid)
1034destNid (SigShutdown) = Nothing
1189destNid (SigShutdown) = Nothing
@@ -1047,7 +1202,7 @@ unClosure closure = do
1047isValidLocalIdentifier :: Identifier -> NC Bool
1202isValidLocalIdentifier :: Identifier -> NC Bool
1048isValidLocalIdentifier ident = do
1203isValidLocalIdentifier ident = do
1049 node <- ask
1204 node <- ask
1050− liftIO . withMVar (localState node) $ \nSt ->
1205+ liftIO . withValidLocalState node $ \nSt ->
1051 case ident of
1206 case ident of
1052 NodeIdentifier nid ->
1207 NodeIdentifier nid ->
1053 return $ nid == localNodeId node
1208 return $ nid == localNodeId node
@@ -1078,6 +1233,9 @@ postMessage pid msg = do
1078throwException :: Exception e => ProcessId -> e -> NC ()
1233throwException :: Exception e => ProcessId -> e -> NC ()
1079throwException pid e = do
1234throwException pid e = do
1080 node <- ask
1235 node <- ask
1236+ -- throwTo blocks until the exception is received by the target thread.
1237+ -- We cannot easily make it happen asynchronpusly because then 'unlink'
1238+ -- semantics would break.
1081 liftIO $ withLocalProc node pid $ \p -> throwTo (processThread p) e
1239 liftIO $ withLocalProc node pid $ \p -> throwTo (processThread p) e
1083withLocalProc :: LocalNode -> ProcessId -> (LocalProcess -> IO ()) -> IO ()
1241withLocalProc :: LocalNode -> ProcessId -> (LocalProcess -> IO ()) -> IO ()
@@ -1085,17 +1243,17 @@ withLocalProc node pid p =
1085 -- By [Unified: table 6, rule missing_process] messages to dead processes
1243 -- By [Unified: table 6, rule missing_process] messages to dead processes
1086 -- can silently be dropped
1244 -- can silently be dropped
1087 let lpid = processLocalId pid in do
1245 let lpid = processLocalId pid in do
1246+ join $ withValidLocalState node $ \vst ->
1088− mProc <- withMVar (localState node) $ return . (^. localProcessWithId lpid)
1247+ return $ forM_ (vst ^. localProcessWithId lpid) p
1091--------------------------------------------------------------------------------
1249--------------------------------------------------------------------------------
1092-- Accessors --
1250-- Accessors --
1093--------------------------------------------------------------------------------
1251--------------------------------------------------------------------------------
1095−links :: Accessor NCState (Map Identifier (Set ProcessId))
1253+links :: Accessor NCState (BiMultiMap Identifier ProcessId ())
1096links = accessor _links (\ls st -> st { _links = ls })
1254links = accessor _links (\ls st -> st { _links = ls })
1098−monitors :: Accessor NCState (Map Identifier (Set (ProcessId, MonitorRef)))
1256+monitors :: Accessor NCState (BiMultiMap Identifier ProcessId MonitorRef)
1099monitors = accessor _monitors (\ms st -> st { _monitors = ms })
1257monitors = accessor _monitors (\ms st -> st { _monitors = ms })
1101registeredHere :: Accessor NCState (Map String ProcessId)
1259registeredHere :: Accessor NCState (Map String ProcessId)
@@ -1104,12 +1262,6 @@ registeredHere = accessor _registeredHere (\ry st -> st { _registeredHere = ry }
1104registeredOnNodes :: Accessor NCState (Map ProcessId [(NodeId, Int)])
1262registeredOnNodes :: Accessor NCState (Map ProcessId [(NodeId, Int)])
1105registeredOnNodes = accessor _registeredOnNodes (\ry st -> st { _registeredOnNodes = ry })
1263registeredOnNodes = accessor _registeredOnNodes (\ry st -> st { _registeredOnNodes = ry })
1107−linksFor :: Identifier -> Accessor NCState (Set ProcessId)
1108−linksFor ident = links >>> DAC.mapDefault Set.empty ident
1110−monitorsFor :: Identifier -> Accessor NCState (Set (ProcessId, MonitorRef))
1111−monitorsFor ident = monitors >>> DAC.mapDefault Set.empty ident
1113registeredHereFor :: String -> Accessor NCState (Maybe ProcessId)
1265registeredHereFor :: String -> Accessor NCState (Maybe ProcessId)
1114registeredHereFor ident = registeredHere >>> DAC.mapMaybe ident
1266registeredHereFor ident = registeredHere >>> DAC.mapMaybe ident
@@ -1136,10 +1288,12 @@ registeredOnNodesFor ident = registeredOnNodes >>> DAC.mapMaybe ident
1136-- * the notifications for typed channels to that process.
1288-- * the notifications for typed channels to that process.
1138-- See https://github.com/haskell/containers/issues/14 for the bang on _v.
1290-- See https://github.com/haskell/containers/issues/14 for the bang on _v.
1291+splitNotif :: (Ord a, Ord v)
1139−splitNotif :: Identifier
1292+ => Identifier
1140− -> Map Identifier a
1293+ -> BiMultiMap Identifier a v
1141− -> (Map Identifier a, Map Identifier a)
1294+ -> (Map Identifier (Set (a,v)), BiMultiMap Identifier a v)
1142−splitNotif ident = Map.partitionWithKey (\k !_v -> ident `impliesDeathOf` k)
1295+splitNotif ident =
1296+ BiMultiMap.partitionWithKeyBy1st (\k !_v -> ident `impliesDeathOf` k)
1144--------------------------------------------------------------------------------
1298--------------------------------------------------------------------------------
1145-- Auxiliary --
1299-- Auxiliary --